Search

The small-strain damping ratio plays a crucial role in assessing the response of soil deposits to earthquake-induced ground motions and general dynamic loading. The damping ratio can theoretically be inverted for after extracting frequency-dependent Rayleigh wave attenuation coefficients from wavefields collected during surface wave testing. However, determining reliable estimates of in-situ attenuation coefficients is much more challenging than achieving robust phase velocity dispersion data, which are commonly measured using both active-source and ambient-wavefield surface wave methods. This paper introduces a new methodology for estimating frequency-dependent attenuation coefficients through the analysis of ambient noise wavefield data recorded by two-dimensional (2D) arrays of surface seismic sensors for the subsequent evaluation of the small-strain damping ratio. The approach relies on the application of an attenuation-specific wavefield conversion and frequency-domain beamforming. Numerical simulations are employed to verify the proposed approach and inform best practices for its application. Finally, the practical efficacy of the proposed approach is showcased through its application to field data collected at a deep, soft soil site in Logan, Utah, USA, where phase velocity and attenuation coefficients are extracted from surface wave data and then simultaneously inverted to develop deep shear wave velocity and damping ratio profiles., Comment: 42 pages, 12 figures

This paper documents a comprehensive subsurface imaging experiment using stress waves in Newberry, Florida, at a site known for significant spatial variability, karstic voids, and underground anomalies. The experiment utilized advanced sensing technologies, including approximately two kilometers of distributed acoustic sensing (DAS) fiber optic cable, forming a dense 2D array of 1920 channels, and a 2D array of 144 three-component nodal stations, to sense active-source and passive-wavefield stress waves. The active-source data was generated using a vibroseis shaker truck and impact sources, and it was simultaneously sensed by both the DAS and the nodal stations. The vibroseis truck was used to excite the ground in the three directions at 260 locations inside and outside the instrumented array, while the impact sources were used at 268 locations within the instrumented array. The passive-wavefield data recorded using the nodal stations comprised 48 hours of ambient noise collected over a period of four days in four twelve-hour time blocks. Meanwhile, the passive wavefield data collected using DAS consisted of four hours of ambient noise recordings. This paper aims to provide a comprehensive overview of the testing site, experiment layout, the DAS and nodal station acquisition parameters, implemented processing steps, and potential use cases of the dataset. While potential use cases, such as surface wave testing, full waveform inversion, and ambient noise tomography, are discussed relative to example data, the focus of this paper is on documenting this unique dataset rather than on processing the data for detecting anomalies or generating subsurface 2D/3D imaging results. The raw and processed data, along with detailed documentation of the experiment and Python tools to aid in visualizing the DAS dataset have been archived and made publicly available on DesignSafe under project PRJ-3521., Comment: 33 pages, 12 figures, dataset paper

This work proposes the AutoHVSR algorithm that allows for fully-automated processing of horizontal-to-vertical spectral ratio (HVSR) measurements, including those with zero, one, or multiple clear resonances. The AutoHVSR algorithm integrates robust signal processing and computational methods with state-of-the-art machine-learning models trained using a diverse dataset of 1109 HVSR measurements. The AutoHVSR algorithm demonstrates excellent performance by correctly determining the number of HVSR resonances for 1099 of the 1109 HVSR measurements (>99%) and predicting the mean resonant frequency of the correctly identified resonances with a root mean square error (RMSE) of 0.05 Hz. Furthermore, the AutoHVSR algorithm was able to produce these predictions in 13 minutes (including HVSR processing time) compared to the 30 hours required for traditional processing (a speed up of 138). The AutoHVSR algorithm is further demonstrated on a challenging dataset from Canterbury, New Zealand that included many HVSR curves with multiple and/or ambiguous resonances. The AutoHVSR algorithm was capable of correctly determining the number of HVSR resonances for 113 of the 129 HVSR measurements (>87%) and predicting the mean resonant frequency of the correctly identified resonances with a RMSE of 0.10 Hz. The AutoHVSR algorithm produced these predictions under 2 minutes (including HVSR processing time) compared to the 4 hours required for traditional processing (a speed up of 120). Finally, while the AutoHVSR algorithm was developed using microtremor measurements where the horizontal components were combined using the geometric mean, it is shown to extend without modification to microtremor HVSR measurements where the two horizontal components are rotated azimuthally and to HVSR measurements from earthquake recordings. The AutoHVSR algorithm has been made publicly available in v0.3.0 of HVSRweb., Comment: 25 pages, 11 figures

There is a growing need to characterize the engineering material properties of the shallow subsurface in three-dimensions for advanced engineering analyses. However, imaging the near-surface in three-dimensions at spatial resolutions required for such purposes remains in its infancy and requires further study before it can be adopted into practice. To enable and accelerate research in this area, we present a large subsurface imaging dataset acquired using a dense network of three-component (3C) nodal stations acquired in 2019 at the Garner Valley Downhole Array (GVDA) site. Acquisition of this dataset involved the deployment of 196, 3C nodal stations positioned on a 14 by 14 grid with a 5-m spacing. The array was used to acquire active-source data generated by a vibroseis truck and an instrumented sledgehammer, and passive-wavefield data containing ambient noise. The active-source acquisition included 66 vibroseis and 209 instrumented sledgehammer source locations. Multiple source impacts were recorded at each source location to enable stacking of the recorded signals. The active-source recordings are provided in terms of both raw, uncorrected units of counts and corrected engineering units of meters per second. For each source impact, the force output from the vibroseis or instrumented sledgehammer was recorded and is provided in both raw counts and engineering units of kilonewtons. The passive-wavefield data includes 28 hours of ambient noise recorded over two night-time deployments. The dataset is shown to be useful for active-source and passive-wavefield three-dimensional imaging as well as other subsurface characterization techniques, which include horizontal-to-vertical spectral ratios (HVSRs), multichannel analysis of surface waves (MASW), and microtremor array measurements (MAM)., Comment: 19 pages, 12 figures

We present a novel method of using explainability techniques to design physics-aware neural networks. We demonstrate our approach by developing a convolutional neural network (CNN) for solving an inverse problem for shallow subsurface imaging. Although CNNs have gained popularity in recent years across many fields, the development of CNNs remains an art, as there are no clear guidelines regarding the selection of hyperparameters that will yield the best network. While optimization algorithms may be used to select hyperparameters automatically, these methods focus on developing networks with high predictive accuracy while disregarding model explainability (descriptive accuracy). However, the field of Explainable Artificial Intelligence (XAI) addresses the absence of model explainability by providing tools that allow developers to evaluate the internal logic of neural networks. In this study, we use the explainability methods Score-CAM and Deep SHAP to select hyperparameters, such as kernel sizes and network depth, to develop a physics-aware CNN for shallow subsurface imaging. We begin with a relatively deep Encoder-Decoder network, which uses surface wave dispersion images as inputs and generates 2D shear wave velocity subsurface images as outputs. Through model explanations, we ultimately find that a shallow CNN using two convolutional layers with an atypical kernel size of 3x1 yields comparable predictive accuracy but with increased descriptive accuracy. We also show that explainability methods can be used to evaluate the network's complexity and decision-making. We believe this method can be used to develop neural networks with high predictive accuracy while also providing inherent explainability., Comment: 26 pages, 14 figures, 4 tables

Distributed acoustic sensing (DAS) is a relatively new technology for recording stress wave propagation, with promising applications in both engineering and geophysics. DAS's ability to simultaneously collect high spatial resolution data over long arrays suggests that it is especially well suited for near-surface imaging applications such as 2D MASW (multi-channel analysis of surface waves). 2D MASW aims to produce a pseudo-2D cross-section of shear-wave velocity (Vs) for the purpose of identifying and characterizing subsurface layering and anomalies. These cross-sections are produced by interpolating numerous 1D Vs profiles extracted from overlapping sub-arrays along the testing alignment. When using traditional seismic equipment, these sub-arrays are typically collected in a roll-along configuration, where the equipment is moved along the alignment at some predetermined sub-array interval. DAS does not have the same limitations, as data from all shot locations are simultaneously recorded along the entire length of the DAS array. This alleviates the requirements to pre-determine sub-array length and interval during the acquisition stage and allows for multiple geometries to be investigated during the processing stage. The present study utilizes DAS data to evaluate the effects of sub-array length on 2D MASW results at a well-characterized field test site. We organize the DAS waveforms into multiple sets of overlapping MASW sub-arrays of differing lengths, allowing for direct comparison of the derived results at the same site. We show that the length of the individual MASW sub-arrays has a significant effect on the resolution of the resulting cross-sections as well as the resolved location of large impedance contrasts at our study site and evaluate those locations compared to invasive testing., Comment: 32 pages, 12 figures

Wellington's port (CentrePort) experienced significant damage from the $M_w$ 7.8 Kaik\=oura earthquake as a result of soil liquefaction, lateral spreading, and shaking-induced damage to structures. To investigate these ill effects, and propose mitigation measures to prevent similar damage in future earthquakes, there was a need to quantify the variations in the depth to bedrock, shear stiffness, and fundamental site period ($T_0$) across the port. In order to characterize $T_0$ and develop shear wave velocity (Vs) profiles for use in seismic site response analyses, horizontal-to-vertical (H/V) spectral ratio measurements and active-source and passive-wavefield surface wave testing (i.e., MASW and MAM, respectively) were performed across the port. A site period map developed from 114 H/V spectral ratio measurements indicates several areas of rapidly changing, complex subsurface structure. Deep (200-plus meters) Vs profiles developed at six reference locations across the port were used to estimate the depth to soft (Vs $>$ 760 m/s) and hard (Vs $>$ 1500 m/s) rock. $T_0$ estimates from H/V spectral ratio measurements ($T_{0,H/V}$) at the six reference locations are shown to be related to the depth of hard rock based on linear viscoelastic transfer functions calculated from Vs profiles truncated at several depths. $T_{0,H/V}$ measurements at two ground motion stations near the port are also shown to be in reasonably good agreement with predominant periods of maximum spectral amplification recorded during both the 2016 Kaik\=oura and 2013 Cook Strait earthquakes, despite these sites also being effected by soil nonlinearity and potential 3D basin edge effects., Comment: Main: 21 pages, 11 figures; Appendix: 8 pages, 6 figures; Final accepted manuscript of Vantassel et al. (2018) published in Bulletin of the Seismological Society of America, released in compliance with Seismological Society of America's green open access policy

This paper presents a frequency-velocity convolutional neural network (CNN) for rapid, non-invasive 2D shear wave velocity (Vs) imaging of near-surface geo-materials. Operating in the frequency-velocity domain allows for significant flexibility in the linear-array, active-source experimental testing configurations used for generating the CNN input, which are normalized dispersion images. Unlike wavefield images, normalized dispersion images are relatively insensitive to the experimental testing configuration, accommodating various source types, source offsets, numbers of receivers, and receiver spacings. We demonstrate the effectiveness of the frequency-velocity CNN by applying it to a classic near-surface geophysics problem, namely, imaging a two-layer, undulating, soil-over-bedrock interface. This problem was recently investigated in our group by developing a time-distance CNN, which showed great promise but lacked flexibility in utilizing different field-testing configurations. Herein, the new frequency-velocity CNN is shown to have comparable accuracy to the time-distance CNN while providing greater flexibility to handle varied field applications. The frequency-velocity CNN was trained, validated, and tested using 100,000 synthetic near-surface models. The ability of the proposed frequency-velocity CNN to generalize across various acquisition configurations is first tested using synthetic near-surface models with different acquisition configurations from that of the training set, and then applied to experimental field data collected at the Hornsby Bend site in Austin, Texas, USA. When fully developed for a wider range of geological conditions, the proposed CNN may ultimately be used as a rapid, end-to-end alternative for current pseudo-2D surface wave imaging techniques or to develop starting models for full waveform inversion., Comment: 34 pages, 13 figures, 2 tables

We rigorously assess the potential for extracting high-resolution, multi-mode surface wave dispersion data from distributed acoustic sensing (DAS) measurements using active-source multichannel analysis of surface waves (MASW). We have utilized both highly-controlled, broadband vibroseis shaker trucks and more-variable, narrow-band sledgehammer sources to excite the near surface, and compare the DAS-derived dispersion data obtained from both source types directly with concurrently acquired traditional geophone-derived dispersion data. We find that the differences between the two sensing approaches are minimal and well within the dispersion uncertainty bounds associated with each individual measurement type when the following conditions are met for DAS: (a) a tight-buffered or strain-sensing fiber-optic cable is used, (b) the cable is buried in a shallow trench to enhance coupling, and (c) short gauge lengths and small channel separations are used. We also show that frequency-dependent normalization of the dispersion image following MASW processing removes the effects of scaling, integration, and differentiation on the measured waveforms, thereby allowing nearly identical dispersion data to be extracted from geophone waveforms and DAS waveforms without requiring them to first be converted into equivalent units. We demonstrate that the short wavelength DAS dispersion measurements are limited by both the gauge length and the more commonly considered channel separation. We further show that it is possible to extract essentially equivalent surface wave dispersion data from seismic measurements made using a traditional geophone array or two different DAS cables. Finally, we show that shear wave velocity profiles recovered from the DAS data using an uncertainty-consistent, multi-mode inversion agree favorably with cone penetration tests performed at the site., Comment: 22 pages, 12 figures

This article extends a recently proposed heterogeneous data assimilation technique for site characterization to estimate compression and shear wave velocity (Vp and Vs, respectively) and damping at Treasure Island and Delaney Park downhole arrays. The adopted method is based on the joint inversion of earthquake acceleration time series and experimental surface wave dispersion data, and including physical constraints to improve the inverse problem's well-posedness. We first use synthetic data at these two sites to refine the proposed approach and then apply the refined algorithm to real data sets available at the Treasure Island and Delaney Park downhole arrays. The joint inversion results show that the estimated Vs and Vp profiles are in very good agreement with measured profiles at these two sites. Our synthetic and real data experiment results suggest that Vp estimation from inversion at downhole arrays can be improved by integrating the water table depth information or the higher modes of the Rayleigh wave dispersion data. Depending on the site complexity, water table information can also help reduce uncertainties associated with damping estimation. In the last part of this article, we compare the performance of the inverted profiles to other methods used to incorporate spatial variability and wave scattering effects in 1D ground response analysis (GRA). The comparisons show that the joint inversion-based Vs and Vp profiles and damping ratios estimated in this article can effectively integrate the effects of spatial variability and wave scattering into 1D GRAs, especially at the Delaney Park downhole array, which is classified as a poorly modeled site using traditional 1D GRA.

The horizontal-to-vertical spectral ratio (HVSR) method has become an increasingly popular tool for developing a quick and reliable estimate of a site's fundamental natural frequency ($f_0$). This paper presents HVSRweb, an open-source, web-based application for performing HVSR calculations in a convenient, reliable, and statistically-consistent manner. HVSRweb allows the user to upload three-component ambient noise records and perform the HVSR calculation in the cloud, with no installation required. The HVSR calculation can be performed using a single combined horizontal component (e.g., geometric-mean, squared-average) or multiple rotated horizontal components to investigate azimuthal variability. HVSRweb can reject spurious time windows using an automated frequency-domain window-rejection algorithm, removing the need for subjective and time-consuming user interaction. It also facilitates the use of lognormal statistics for $f_0$, allowing for a consistent statistical framework capable of quantifying measurement uncertainty in terms of either frequency or its reciprocal, period. In addition, HVSRweb presents the opportunity to rapidly incorporate new developments in processing and quantification of uncertainty as they emerge, and encourages standardization of HVSR processing across users while avoiding the challenges associated with traditional approaches that require the user to perform regular updates to keep pace with new developments., Comment: 10 pages, 5 figures, accepted in the International Foundations Congress and Equipment Expo (IFCEE) 2021, to try HVSRweb for yourself go to https://hvsrweb.designsafe-ci.org/

Non-invasive subsurface imaging using full waveform inversion (FWI) has the potential to fundamentally change engineering site characterization by enabling the recovery of high resolution 2D/3D maps of subsurface stiffness. Yet, the accuracy of FWI remains quite sensitive to the choice of the initial starting model due to the complexity and non-uniqueness of the inverse problem. In response, we present the novel application of convolutional neural networks (CNNs) to transform an experimental seismic wavefield acquired using a linear array of surface sensors directly into a robust starting model for 2D FWI. We begin by describing three key steps used for developing the CNN, which include: selection of a network architecture, development of a suitable training set, and performance of network training. The ability of the trained CNN to predict a suitable starting model for 2D FWI was compared against other commonly used starting models for a classic near-surface imaging problem; the identification of an undulating, two-layer, soil-bedrock interface. The CNN developed during this study was able to predict complex 2D subsurface images of the testing set directly from their seismic wavefields with an average mean absolute percent error of 6%. When compared to other common approaches, the CNN approach was able to produce starting models with smaller seismic image and waveform misfits, both before and after FWI. The ability of the CNN to generalize to subsurface models which were dissimilar to the ones upon which it was trained was assessed using a more complex, three-layered model. While the predictive ability of the CNN was slightly reduced, it was still able to achieve seismic image and waveform misfits comparable to the other commonly used starting models. This study demonstrates that CNNs have great potential as a tool for developing good starting models for FWI ..., Comment: 29 pages, 16 figures, submitted to Geophysics

Non-invasive surface wave methods have become a popular alternative to traditional invasive forms of site-characterization for inferring a site's subsurface shear wave velocity (Vs) structure. The advantage of surface wave methods over traditional forms of site characterization is that measurements made solely at the ground surface can be used routinely and economically to infer the subsurface structure of a site to depths of engineering interest (20-50 m), and much greater depths (>1 km) in some special cases. However, the quantification and propagation of uncertainties from surface wave measurements into the Vs profiles used in subsequent engineering analyses remains challenging. While this has been the focus of much work in recent years, and while considerable progress has been made, no approach for doing so has been widely accepted, leading analysts to either address the propagation of uncertainties in their own specialized manner or, worse, to ignore these uncertainties entirely. In response, this paper presents an easy-to-implement, effective, and verifiable method for developing uncertainty-consistent Vs profiles from inversion of surface wave dispersion data. We begin by examining four approaches presented in the literature for developing suites of Vs profiles meant to account for uncertainty present in the measured dispersion data. These methods are shown to be deficient in three specific ways. First, all approaches are shown to be highly sensitive to their many user-defined inversion input parameters, making it difficult/impossible for them to be performed repeatedly by different analysts. Second, the suites of inverted Vs profiles, when viewed in terms of their implied theoretical dispersion data, are shown to significantly underestimate the uncertainty present in the experimental dispersion data, though some may appear satisfactory when viewed purely qualitatively..., Comment: 25 pages, 15 figures; fixed typos, included doi, no changes made to figures or discussion

SWinvert is a workflow developed at The University of Texas at Austin for the inversion of surface wave dispersion data. SWinvert encourages analysts to investigate inversion uncertainty and non-uniqueness in shear wave velocity (Vs) by providing a systematic procedure and open-source tools for surface wave inversion. In particular, the workflow enables the use of multiple layering parameterizations to address the inversion's non-uniqueness, multiple global searches for each parameterization to address the inverse problem's non-linearity, and quantification of Vs uncertainty in the resulting profiles. To encourage its adoption, the SWinvert workflow is supported by an open-source Python package, SWprepost, for surface wave inversion pre- and post-processing and an application on the DesignSafe-CyberInfracture, SWbatch, that enlists high-performance computing for performing batch-style surface wave inversion through an intuitive and easy-to-use web interface. While the workflow uses the Dinver module of the popular open-source Geopsy software as its inversion engine, the principles presented can be readily extended to other inversion programs. To illustrate the effectiveness of the SWinvert workflow and to develop a set of benchmarks for use in future surface wave inversion studies, synthetic experimental dispersion data for 12 subsurface models of varying complexity are inverted. While the effects of inversion uncertainty and non-uniqueness are shown to be minimal for simple subsurface models characterized by broadband dispersion data, these effects cannot be ignored in the Vs profiles derived for more complex models with band-limited dispersion data. The SWinvert workflow is shown to provide a methodical procedure and a powerful set of tools for performing rigorous surface wave inversions and quantifying the uncertainty in the resulting Vs profiles., Comment: 25 pages, 11 figures